城市的快速发展及其功能需求的日益变化使得土地利用日趋复杂与多元,如何自动、高效地获取精细的城市土地利用信息成为当前亟需解决的问题。针对以深度学习为代表的数据驱动方法在可解释性弱与样本依赖性强的局限,本文提出一种基于语义的知识驱动城市土地利用分类方法。该方法融合遥感影像、POI等多源数据与产品,首先基于EAGLE矩阵对土地利用类型进行语义重构,解析出本体基元；进而提取众源数据特征、遥感指数与形态学特征,以匹配基元的属性特征,并通过基元组合构建本体模型。在此基础上,提出一种融合“光谱+路网”的分割策略,生成同质地块作为本体推理的实例。在构建本体模型过程中,通过基元与数据特征之间的关系建立约束规则,最终依据基元组合规则推理出每个地块的土地利用类型。为验证方法有效性,选取土地利用高度复杂的加拿大多伦多市中心作为实验区,结果表明：基于语义的知识驱动分类结果总体精度达到87.02%,Kappa系数为0.84。该方法不仅实现了高精度分类,同时借助光谱与路网的分割使地块边界更符合实际地表情况,为城市土地利用识别提供了新思路,适用于城市区域土地利用的精准识别与监测。

The rapid development of cities and the ever-changing functional demands have made land use increasingly complex and diverse. How to automatically and efficiently obtain detailed urban land use information has become an urgent problem to be solved at present. In view of the limitations of data-driven methods represented by deep learning, such as weak interpretability and strong sample dependence, this paper proposes a semantic-based knowledge-driven urban land use classification method. This method integrates multi-source data and products such as remote sensing images, POI, AOI, OSM, LCZ and WSF. Firstly, it conducts semantic reconstruction of land use types based on the EAGLE matrix and parses the ontology primitives. Furthermore, the features of crowdsource data, remote sensing indices and morphological features are extracted to match the attribute features of primitives, and the ontology model is constructed through the combination of primitives. On this basis, a segmentation strategy integrating "spectrum + road network" is proposed to generate homogeneous plots as instances of ontology reasoning. In the process of constructing the ontology model, constraint rules are established through the relationship between primitives and data features, and finally the land use type of each plot is inferred based on the combination rules of primitives. To verify the effectiveness of the method, the downtown area of Toronto, Canada, where land use is highly complex, was selected as the experimental area. The results show that the overall accuracy of the semantic-based knowledge-driven classification results reaches 87.02%, and the Kappa coefficient is 0.84. This method not only achieves high-precision classification, but also makes the boundaries of plots more in line with the actual surface conditions by means of spectral and road network segmentation, providing a new idea for the identification of urban land use. It is suitable for the precise identification and monitoring of land use in urban areas.